ARG-ST as a head feature

[Martin Jansche]

On Sat, 20 Jan 2001, Karel Oliva wrote:

> EITHER locality is not a very persuasive argument

Locality does not have any impact on the expressiveness of the
framework.  If one only wanted grammars that are strictly local, this
could easily be achieved.  Perhaps not elegantly, but that's not the
point.

I've come to view locality as a property that simplifies reasoning
about (non-circular) feature structures.  If all constraints are
local, one can use simple induction to prove stuff.  If not, perhaps
complete induction might work, depending on the nature of the
constraints (e.g., if they only "inspect" structures, rather than
looking all over the place), but not in general.  So if you've written
a grammar and are trying to prove it to conform to a certain
specification, you might be better off if your grammar did obey
locality.  However, I currently fail to see any other rationales for
enforcing locality: elegance does not count unless rigorously defined,
and empirical consequences are not at stake.

- martin jansche